Professional Documents
Culture Documents
Musical Articulation in The Organ
Musical Articulation in The Organ
Musical Articulation in The Organ
Jesper Jerkert
ABSTRACT
In organ playing the only musical parameters immediately available for expression are changes of the tone
onsets and offsets. While interonset intervals (IOIs) have been studied extensively in performances on several
instruments, the offset timings are largely unknown. For analysis of musical articulation the extractions of
onsets as well as offsets are required.
In this study, the tone onsets and offsets were measured in seven monophonic Bach organ fugue openings,
each of which was played by three, four, or five organists sampled from commercially available recordings.
The articulation for each tone was defined as the ratio between the tone duration and the IOI. The articula-
tions, IOIs, and relative IOIs of all tones were calculated, as well as a number of inter- and intrapersonal cor-
relation coefficients. The relative IOI is a measure of the local tempo deviation. Articulatory strategies were
analysed using an explorative (not hypothesis-driven) approach.
Quantitatively, the mean articulation was 0.86, with an average standard deviation of 0.15. The mean cor-
relation coefficient between performers’ articulation in the same piece was 0.44, which is almost as high as for
the relative IOI (0.46). The mean correlation between articulation and relative IOI was –0.26 (95% CI –0.37 -
–0.16). In the studied material lengthening of relative IOI thus typically coincides with shorter (more staccato-
like) articulation.
Qualitatively, some performances were articulated binarily, i.e. each tone is long or short but not inter-
mediate. Other articulation strategies enhance the inherent phrasing or matrical structure. Different articulation
strategies may well be mutually contradictory. Therefore, possible rules in a model for musical articulation
should not be applied indiscriminately, but should be applied in groups falling under different overall strat-
egies.
1. INTRODUCTION
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-1
see [4]. For example, when performers are asked to play a piece first in a neutral way and then in a specified
expressive way (e.g. “happy” or “sad”), the articulation is changed, as are many other factors.
The German musicologist Hermann Keller is an important person in this area. In 1925 he published Die
musikalische Artikulation insbesondere bei Joh. Seb. Bach (“Musical Articulation, Especially in Joh. Seb.
Bach”, no English translation available) [5]. Thirty years later, he completely rewrote the book and published
it with the title Phrasierung und Artikulation (Phrasing and Articulation, English translation 1965). As for the
distinction between articulation and phrasing – sometimes used synonymously by musicians and musicologists
– Keller was clear-cut: “[T]he words ‘phrasing’ and ‘articulation’ have basically different meanings: phrasing
is much like the subdivision of thought; its function is to link together subdivisions of musical thought
(phrases) and to set them off from one another; it thus has the same function as punctuation marks in language.
(…) The function of musical articulation, on the other hand, is the binding together or the separation of the
individual notes; it leaves the intellectual content of a melody line inviolable, but it determines its expression”
[6].
It is important to recognize that there is a variety of strategies for making adjacent tones belong together or
for making them seem separated. Tones can be played longer or shorter than their nominal duration. Further-
more, tones can be played with different sound levels, or with different vibrato, or with different tone attacks,
etc. Adjustment of the amount of rest between the tones – the most obvious articulation strategy – is far from
the only possibility.
In this paper I will only deal with methods involving the timing of the tones. There are two such methods.
The first is to alter the relative length of the rest, i.e. what is normally called legato vs. staccato. If one note
ends exactly when the following note starts, the articulation ratio is 1 (one). If one note lasts for only half the
time between its onset and the onset of the next tone, the articulation ratio is 0.5. In other words the
articulation ratio for tone n is (offsetn – onsetn)/IOIn, where IOI is the interonset interval (see e.g. [7]), defined
as the time interval between the onset of a given tone and the onset of the immediately following tone, i.e. IOIn
= onsetn+1 – onsetn. For simplicity, I will use the word “articulation” for “articulation ratio” as defined above.
It should be noted, incidentally, that the term “IOI” is well-established from earlier studies, whereas
“articulation” in the specified sense “(offset – onset)/IOI” perhaps is not. Some researchers, e.g. [8], use the
terms “key overlap time” (KOT) and “key detached time” (KDT) in discussing legato and staccato in piano
playing. The word “articulation” covers them both, being also applicable in other musical instruments.
The second method is to play the tones (or, more to the point, their IOI’s) longer or shorter than their
nominal duration. In other words the tempo is altered locally. This can be expressed in terms of relative IOI. A
relative IOI of 1 (one) means that the duration of the tone is exactly what is to be expected from the nominal
(notated) length given the mean tempo of the piece. For example, if a piece is performed at the mean tempo
120 quarter-notes per minute, we would expect an eighth-note to last for 0.25 s. A certain eighth-note with a
duration of 0.30 s then has the relative IOI 0.30/0.25 = 1.2.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-2
On articulation, Gabrielsson writes that “[a]lmost all the present performances are of the legato type” (p. 99).
He discusses some exceptions to this observation, but there is no evidence that he actually measured the tone
offsets.
Repp made a very large study of “temporal commonalities and differences” in piano playing [11]. Repp
measured the IOIs in no less than 28 recorded performances of Schumann’s “Träumerei”. His focus is on the
behaviours such as ritardandi of structures called “melodic gestures” (two to seven notes long). Repp’s paper
contains many interesting observations. Again, it has nothing to report on articulation, since only the tone
onsets were measured. The introductory part (p. 2546–2548) gives a valuable overview of earlier research.
In three subsequent papers, Repp specifically studied legato and staccato articulation in the piano [12],
[13], [14]. However, the data material for these works were scales, arpeggios, and simple sequences, not “real”
music as in the earlier study [11]. In all three papers the pianists were explicitely asked to play legato or
staccato. Repp’s findings were, for example: KOTs (key overlap times) for successive tones judged to be
optimally legato were greater for high than for low tones [12], [13]; note durations increased significantly as
tempo decreased in both perception and production of staccato [14]. As indicated by the article titles, one of
Repp’s main concerns has been the relationship between perception and production of different articulations.
He claims, e.g., that produced staccato is not necessarily percepted as staccato [14].
One interesting study on piano playing was made by Battel and Fimbianti [15]. They studied the first 16
bars of the Andante movement of Mozart’s G Major Sonata, K. 545. Five near-professional pianists played the
piece nine times each. First they were asked to play it in an “optimal” (preferred) way, then in eight ways
characterized by four pairs of expressive adjectives (bright vs. dark, light vs. heavy, hard vs. soft, and
passionate vs. flat). The researchers performed several statistical tests relating the tone timings to the different
adjectives (“intentions”). As for the articulation, they stated: “A close survey proves that the staccato/legato
degree in each version depends on [harmonic] tension. Particularly, the DRO average of all performance notes
depends on the expressive intention while, for each note, local value depends directly on the harmonic
structure. The performer increases the slur degree when the harmonic tension increases in each different
intention, no matter the global value” (p. 69f). The DRO is explained to be “the ratio between the note value
(key on/off) and the IOI” (p. 69), which appears to be identical to articulation in the present work. (This is
confusing, since DRO in other papers on music performance stands for the pause duration, not the tone
duration.) In other words, Battel and Fimbianti claim that the average articulation is affected by the overall
intention, but also that the individual values are dependent on the harmonic tension of each note. We are told
that the tension for each note was calculated according to a theory due to Lerdahl. This result sounds very
interesting, but since no details are given it is hard to assess the quality of the finding.
Bresin and Battel [8] used the same material for further analysis. They concentrated more on articulation
matters. They found that legato was played with a key overlap ratio which depended on the IOI.
When both onsets and offsets have been measured, as in some of the above studies, the material has been
either scales and arpeggios or music where the composer has already added some phrasing and articulatory
marks. The comprehensive reviews by Alf Gabrielsson [16], [17] confirm this picture; in studies of perform-
ance timing only the onsets seem to have been extracted in most cases.
The present study aims at a rough description of the articulation in one-part melodies played on the organ. I
will report some basic findings. In particular I will investigate the following questions: What is “normal”
articulation in one-part music lacking notated articulation marks, i.e. what is its mean and standard deviation?
How large differences are there between performers in the same piece? Is there any detectable relationship
between articulation and IOI? Between articulation and relative IOI? If so, how could these be explained?
It seems that no one has investigated these questions before, at least not in a quantitative manner. There are
some studies on how to play a predetermined articulation type, as indicated in the preceding section. For
example, Battel and Bresin [8] reports that a typical staccato in the pianoforte has an articulation ratio of
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-3
approximately 0.4 (independent of absolute IOI). Repp [12] reports that a legato in the piano has an articula-
tion ratio of about 1.2. However, there are no reports on the articulation of music without notated articulation
marks. I will also devote some space to finding out whether specific articulation strategies are present in the
pieces analysed.
The reason for analysing organ recordings rather than, say, violin recordings is that the articulation is
straight-forward in the organ; it is only a matter of measuring the onset and offset of each tone. Furthermore,
the organ tone is steady, being (very nearly) equally loud from onset to offset. A possibly complicating factor
is that there is no standard organ. For example, organs might be mechanical or electro-pneumatic. Different
organs do not respond in the same way to a given touch. These things have been ignored in this work. The
rationale for this is the assumption that a skilful organist can produce an intended articulation on almost any
organ (or at least on the organs of the analysed recordings). Some of the preliminary results of this study have
been summarised in [18].
3. METHOD
The material in this study is seven Bach fugue openings for the organ. The material is from BWVs 537, 538,
542, 544, 548, 564, and 578. By “opening” I mean from the beginning of the fugue (dux) to the entrance of the
second voice (comes) or even shorter; the material is thus monophonic. Scores of the fugue themes are pre-
sented in Figure 1. These scores were redrawn from [5]. In BWV 538 the trill of the tone e has been excluded
from analysis. The same is true for the trill f sharp in BWV 548. In both cases the articulation ratios are
ascribed the value 1 (one). There are notated pauses in BWV 564. Here, notes immediately preceding pauses
have been excluded from analysis. This is necessary because it is impossible to calculate their tempo devi-
ations, and hence their relative IOIs.
a)
b)
c)
d)
e)
f)
g)
Figure 1: The seven analysed fugue openings: a) BWV 537, b) BWV 538, c) BWV 542, d) BWV 544, e)
BWV 548, f) BWV 564, g) BWV 578.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-4
Each piece is played by three, four, or five organists that were sampled from commercially available record-
ings. In all, 28 recordings were used. The organists are Hans Fagius, Peter Hurford, Wolfgang Rübsam, Hel-
muth Rilling, and Ton Koopman. See Table 1 for details.
The onset and offset of each tone were estimated through spectrogram inspections. The programs Wave-
Surfer 1.4.5 [19] and Soundswell 4.0 [20] were used to supply spectrograms. Because of the very reverberant
(and very different) recording conditions it was not possible to use an automatic procedure to establish the
onset and offset timings, e.g. by defining them as the moment when the sound level exceeds or falls below a
fixed number. The extraction of timings has thus been made manually, through ocular inspection combined
with repeated playbacks. For each performance the onset/offset timings were input to Matlab files calculating
the articulation of each tone, the relative IOI and some statistical parameters.
As for the search for specific articulation strategies, an explorative approach will be used. This means that
I do not test any predetermined hypotheses, but try to find patterns in the articulation. For example, if I analyse
a piece with this passage appearing in the printed notes, , finding that one performer articulates like
this, , I would guess that it was intentionally done, because it seems musically reasonable. There-
fore it would qualify as a “pattern”. But finding such patterns is, of course, a risky project, since I might
attribute intentions to the performers originating, in reality, from my fantasy.
BWV Performer Recording Year
537 Hans Fagius BIS 397/398 c. 1988
537 Peter Hurford Decca 443485-2 1979
537 Wolfgang Rübsam Naxos 8553150 1988
537 Ton Koopman Teldec 0630-13155-2 1996
538 Hans Fagius BIS 439/440 1989
538 Peter Hurford Decca 443485-2 1979
538 Wolfgang Rübsam Naxos 8553150 1988
538 Ton Koopman Archiv 410999-2 1983
542 Hans Fagius BIS 308/309 c. 1985
542 Peter Hurford Decca 443485-2 1978
542 Wolfgang Rübsam Naxos 8550652 1989
542 Helmuth Rilling Denon 38C37-7039 1974
542 Ton Koopman Hänssler 98182 1986
544 Hans Fagius BIS 397/398 1988
544 Wolfgang Rübsam Naxos 8550652 1989
544 Helmuth Rilling Denon 38C37-7039 1974
544 Ton Koopman Teldec 4509-94458-2 1994
548 Hans Fagius BIS 397/398 1988
548 Peter Hurford Decca 421337-2 1981
548 Wolfgang Rübsam Naxos 8550184 1988
564 Hans Fagius BIS 343/344 1986
564 Peter Hurford Decca 443485-2 1981
564 Wolfgang Rübsam Naxos 8550901 1993
564 Ton Koopman Archiv 410999-2 1983
578 Hans Fagius BIS 439/440 1989
578 Wolfgang Rübsam Naxos 8553135 1995
578 Helmuth Rilling Denon 38C37-7039 1974
578 Ton Koopman Teldec 4509-94458-2 1994
Table 1: Recordings used for analysis.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-5
4. RESULTS
4.1. Tempos
Table 2 shows the tempos chosen by the organists in all performances.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-6
4.2. Articulation
The mean articulation, for all pieces and all performances, was 0.86, with an average standard deviation of
about 0.15. The results for each performance are presented in Table 3. As can be seen, there are some differ-
ences between pieces, which is quite expected because different pieces seem to invite the performers to artic-
ulate differently. Two piece mean values stand out. In BWV 544 the mean value is high, with a small standard
deviation, whereas in BWV 564 the mean value is low, with a high standard deviation.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-7
BWV Pair of performers Art. corr. Piece mean
537 Fagius-Hurford 0.31
537 Fagius-Rübsam 0.70
537 Fagius-Koopman 0.71
0.50
537 Hurford-Rübsam 0.32
537 Hurford-Koopman 0.59
537 Rübsam-Koopman 0.38
538 Fagius-Hurford 0.93
538 Fagius-Rübsam 0.54
538 Fagius-Koopman 0.84
0.63
538 Hurford-Rübsam 0.31
538 Hurford-Koopman 0.82
538 Rübsam-Koopman 0.36
542 Fagius-Hurford 0.22
542 Fagius-Rübsam 0.51
542 Fagius-Rilling 0.34
542 Fagius-Koopman 0.63
542 Hurford-Rübsam 0.12 0.36
542 Hurford-Rilling –0.14
542 Hurford-Koopman –0.01
542 Rübsam-Rilling 0.54
542 Rübsam-Koopman 0.61
542 Rilling-Koopman 0.76
544 Fagius-Rübsam 0.20
544 Fagius-Rilling 0.89
544 Fagius-Koopman 0.01
0.30
544 Rübsam-Rilling 0.42
544 Rübsam-Koopman 0.14
544 Rilling-Koopman 0.17
548 Fagius-Hurford –0.07
548 Fagius-Rübsam 0.34 0.13
548 Hurford-Rübsam 0.11
564 Hurford-Koopman 0.60
564 Hurford-Rübsam 0.52
564 Hurford-Fagius 0.71
0.73
564 Koopman-Rübsam 0.83
564 Koopman-Fagius 0.82
564 Rübsam-Fagius 0.90
578 Fagius-Rübsam 0.40
578 Fagius-Rilling 0.22
578 Fagius-Koopman 0.50
0.33
578 Rübsam-Rilling 0.18
578 Rübsam-Koopman 0.16
578 Rilling-Koopman 0.51
Table 4. Articulation correlations between all performer pairs.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-8
4.2.2. Articulation and IOI
The IOI (or, for clarity, absolute IOI) is the duration of the tone, including the following pause (if any) before
the next tone. The IOI is straightforward to perceive. Relative IOI, on the other hand, is cognitively much more
complicated, since it reflects the deviation from a notated duration, the existence of which the listener must
deduce from the relative values only. Let us therefore first investigate whether we can find any interesting cor-
relation coefficients when comparing absolute IOI and articulation values. Table 5 shows the correlations be-
tween articulation and absolute IOI for each performance.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-9
another articulation (here: shorter) than the short notes, the correlation can easily be very significant, because
the notes are divided into two distinct groups along both the IOI and articulation dimensions.
The most positive correlations are found in BWV 538. Here, too, long and short notes are articulated
differently, but the other way around; long notes are played more legato, and short notes more staccato. Why?
I think the reason is that the short notes (quarter-notes) generally occur on weak beats and are followed by
leaps. They therefore function as up-beats.
For reference, all data are presented in Figure 3.
1.5
1
Articulation
0.5
0
0 0.5 1 1.5 2 2.5
IOI
Figure 3. IOI vs. articulation for all data, i.e. all performances of all pieces. IOI is measured in seconds.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-10
There were no great differences between pieces, except that BWV 564 stands out with an unussualy negative
mean.
BWV Performer Corr. art.-rel. IOI Piece mean
537 Fagius –0.36
537 Hurford 0.21 –0.11
537 Rübsam –0.34
537 Koopman 0.07
538 Fagius –0.28
538 Hurford 0.14 –0.22
538 Rübsam –0.63
538 Koopman –0.11
542 Fagius –0.53
542 Hurford –0.02
542 Rübsam –0.51 –0.34
542 Rilling –0.02
542 Koopman –0.61
544 Fagius –0.58
544 Rübsam 0.07 –0.24
544 Rilling –0.51
544 Koopman 0.06
548 Fagius 0.09
548 Hurford –0.24 –0.20
548 Rübsam –0.44
564 Fagius –0.66
564 Hurford –0.08 –0.50
564 Rübsam –0.74
564 Koopman –0.52
578 Fagius –0.26
578 Rübsam –0.47 –0.19
578 Rilling –0.04
578 Koopman 0.03
Table 6. Correlations between articulation and relative IOI for all performances.
Another way of handling this material is to look at the mean value for each performer. Table 7 shows the
correlations from Table 6 averaged for each performer.
Performer (no.
Mean correlation
of performan-
art.-rel. IOI
ces)
Fagius (7) –0.37
Hurford (5) 0.00
Rübsam (7) –0.44
Rilling (3) –0.19
Koopman (6) –0.18
Table 7. Mean value for correlations between articulation and relative IOI for all performers.
Table 7 is interesting information. Of course, the data set is too little to establish the calculated values as
“typical” for these performers (this is especially true for Rilling, with only three performances). But the set is
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-11
sufficient for the assessment that there seems to be a variability in the articulation/relative IOI correlation
between performers.
The whole material is presented in Figure 4.
1.5
1
Articulation
0.5
0
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5
Relative IOI
Figure 4. Relative IOI vs. articulation for all data, i.e. all performances of all pieces.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-12
1.5
Articulation Rubsam 1
0.5
0
0 0.5 1 1.5
Articulation Fagius
Figure 5. Articulation in BWV 537 for Fagius and Rübsam. The two outliers closest to the bottom right
corner are the two disputed eighth-notes D and E flat at the end of the theme.
Figure 6. Articulation in BWV 537 for Fagius, Rübsam, and Koopman. Tones for which the articulation
is less than 0.8 (for Fagius and Rübsam) or less than 0.5 (for Koopman) are marked staccato, other notes
are marked portato. The two tones marked both ways were played staccato by Rübsam but legato by
Fagius and Koopman.
Hurford uses roughly the same articulation, but he is more extreme, reaching articulations below 0.6 on
three occations. At the same time he plays with the most stable tempo of the three performers, his relative IOI
never exceeding ±10% (see Figure 8, top). It seems resonable to say that Hurford compensates his moderate
tempo deviations with more expressive articulation; but this can only be an intelligent guess given the small
data sample. Figure 7 shows Hurford’s articulation in musical notes.
Figure 7. Hurford’s articulation in BWV 537. Staccato dots indicate articulations of below 0.6. All
other tones have articulations above 0.9.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-13
Figure 8 shows all data for articulation and relative IOI.
1.4
0.8
0.6
0 0.5 1 1.5 2 2.5 3 3.5
Whole tones
Fagius
Hurford
1.5 Rubsam
Koopman
Articulation
0.5
Figure 8: Relative IOI (top) and articulation (bottom) for all four performances of BWV 537.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-14
1.4
1.2
Relative IOI
1
0.8
0.6
0 1 2 3 4 5 6 7
Whole tones
Fagius
Hurford
1.5 Rubsam
Koopman
Articulation
0.5
0 1 2 3 4 5 6 7
Whole tones
Figure 9: Relative IOI (top) and articulation (bottom) for all four performances of BWV 538.
Figure 10. Articulation in BWV 542 for Fagius. Staccato dots indicate articulations of 0.56–0.79,
portato strokes indicate articulations of 1.03–1.05, neither dots nor strokes indicate articulations of
0.81–0.98.
Turning to Hurford, it is not as easy to find distinct articulation types. Almost all articulation ratios are
between 0.86 and 1.15. Four notes have articulation values below 0.76. These four notes are marked with
staccato dots in Figure 11. They are not randomly scattered but are the last two notes in a recurrent group of
four. All other notes have articulation values of 0.86 or more.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-15
Figure 11. Articulation in BWV 542 for Hurford. Staccato dots indicate articulations of less than 0.76.
All other notes have articulations of 0.86 or more.
According to Table 3, in Koopman has a mean articulation of 0.70, whereas Rilling has 0.93. Still, their
articulation profiles are very similar, as shown in Figure 12. They have both adopted a binary type of articula-
tion; either short (for Koopman 0.37–0.57, for Rilling 0.57–0.75) or long (for Koopman 0.73–1.04, for Rilling
1.02–1.20). Since they have chosen the same type of articulation for all tones, with only one or two exceptions,
the points in Figure 12 are gathered in two distinct groups. What is the explanation of the striking similarity
found in Koopman and Rilling? The answer is simple: All long notes (eighth-notes) are played short, while all
short notes (sixteenth-notes) are played long (note the highly negative correlations for Koopman and Rilling in
BWV 542 in Table 5).
1.5
1
Articulation Rilling
0.5
0
0 0.5 1 1.5
Articulation Koopman
Figure 12. Articulation in BWV 542 for Koopman and Rilling. The binary character is clear in both
performers, but it is most accentuated in Rilling.
Rübsam’s articulation is less distinct. Nonetheless, looking at Figure 13 (lower panel), Rübsam uses ap-
proximately the same articulation as Koopman and Rilling, the difference being that Rübsam’s is not binary
but more blurred.
All data are shown in Figure 13.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-16
1.4
1.2
Relative IOI
1
0.8
0.6
0 0.5 1 1.5 2 2.5
Whole tones
Fagius
Hurford
1.5 Rubsam
Rilling
Koopman
Articulation
0.5
Figure 13: Relative IOI (top) and articulation (bottom) for all five performances of BWV 542.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-17
1.4
1.2
Relative IOI 1
0.8
0.6
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Whole tones
Fagius
Rubsam
1.5 Rilling
Koopman
Articulation
0.5
Figure 14: Relative IOI (top) and articulation (bottom) for all four performances of BWV 544.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-18
1.4
1.2
Relative IOI 1
0.8
0.6
0 0.5 1 1.5 2 2.5 3 3.5 4
Whole tones
1.5 Fagius
Hurford
Rubsam
Articulation
0.5
Figure 15: Relative IOI (top) and articulation (bottom) for all three performances of BWV 548. The trill
has been assigned the articulation 1 (one) in all performances.
Figure 16. Articulation in BWV 564 for Hurford and Fagius. Staccato dots indicate articulations below
0.5 (for Hurford) or below 0.65 (for Fagius). Two tones are disputed.
Koopman’s and Rübsam’s articulations are similar to Hurford’s and Fagius’, but with even more tones
made staccato. In particular, the fourth beat of each bar is played staccato. This was not the case with Hurford
and Fagius. Koopman and Rübsam have very different “standard articulation” levels, their average articulation
values being 0.65 and 0.86, respectively. Still, we see here another example of great similarities found in per-
formers with very different average articulations. Figure 17 shows their articulations plotted against one
another, Figure 18 shows the notated result.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-19
1.5
1
Articulation Rubsam
0.5
0
0 0.5 1 1.5
Articulation Koopman
Figure 18. Articulation in BWV 564 for Koopman and Rübsam. Staccato dots indicate articulations
below 0.6 (for Koopman) or below 0.85 (for Rübsam).
Figure 19 shows all data.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-20
1.4
1.2
Relative IOI
1
0.8
0.6
0 0.5 1 1.5 2 2.5
Whole tones
Fagius
Hurford
1.5 Rubsam
Koopman
Articulation
0.5
Figure 19: Relative IOI (top) and articulation (bottom) for all four performances of BWV 564. Please
observe that notes followed by pauses have been excluded.
Figure 20. Articulation in BWV 578 for Rübsam. Staccato dots indicate articulation below 0.9.
Figure 21. Articulation in BWV 578 for Fagius. Staccato dots indicate articulation below 0.85
Rilling shows an almost uniform legato articulation throughout the excerpt. Scrutinizing his few staccato
tones reveals that they appear at the end of bars 2 and 3. I believe this is Rilling’s way of marking the phrase
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-21
ends; he thus sees bars 1-2 as one phrase, bar 3 as one phrase, and bars 4-5 (or possibly even longer) as one
phrase. Figure 22 shows Rilling’s articulation in musical notes.
Figure 22. Articulation in BWV 578 for Rilling. Staccato dots indicate articulation below 0.8.
Koopman’s articulation is crisper. He plays staccato on all eighth-notes followed by leaps, plus the last
note of the second bar. See Figure 23.
Figure 23. Articulation in BWV 578 for Koopman. Staccato dots indicate articulation below 0.715.
Figure 24 shows all data.
1.4
1.2
Relative IOI
0.8
0.6
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Whole tones
Fagius
Rubsam
1.5 Rilling
Koopman
Articulation
0.5
Figure 24: Relative IOI (top) and articulation (bottom) for all four performances of BWV 578.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-22
5. DISCUSSION AND CONCLUSIONS
This investigation has given some quantitative results. First, the mean articulation was 0.86, with a standard
deviation of 0.15 across the whole material. It is not easy to know how strongly influenced is the result from
the type of instrument (organ) or from the type of music (Bach fugue openings). Articulation similarity be-
tween performers of the same piece was almost as high (mean correlation 0.44) as the relative IOI similarity
(mean correlation 0.46). The difference is too small to be statistically significant, but if it is real I believe it
can be explained by the relative IOI being not only an articulation strategy, but most of all a phrasing strategy.
Relative IOI adjustments act also on a larger scale, making them more robust.
Secondly, in most performances, the correlation was negative between articulation and relative IOI. Al-
though there were differences between pieces (Table 6, right column), they were not greater than those be-
tween performers (Table 7). Therefore, as an approximation of the first order, we could say that articulation
and relative IOI are inversely related, roughly speaking.
A general result of more qualitative character is that performers may have different average articulation
but still share the same articulatory strategy. Thus an articulation of 0.75 could be staccato for one performer,
but legato for another. This might perhaps not be surprising, but it is nevertheless a basic result to keep in
mind in all discussions on articulation.
Up-beat tones are often played short in the studied material. It is possible that this is a rule that applies to a
vast majority of all cases.
Another result is the occurence of binary articulation. This term means that the performer articulated each
tone either short or long. The most clear examples were found in BWV 542 and BWV 564. Both these themes
are rather fast and are made out of only two note-values. The existence of binary articulation is interesting
from a computer-control point of view, for it could facilitate successful simulation of realistic articulation. It
would only be a matter of determining for each tone whether it is short or long. Within the two articulation
groups the exact articulation values could be computed by adding or subtracting small random values so that
not all tones in the same group (short or long) will get exactly the same articulation.
Other articulation strategies enhance the phrasing or the metrical structure (with more staccato on weak
beats). Some strategies seem to be mutually exclusive.
All in all, we have ended up with a bundle of diverse results. To be fair, they do point at any clearly domi-
nant factor always determining the articulation.
Therefore, when modelling musical articulation, e.g. in order to implement a rule system for computers,
one should probably not try to apply all conceivable rules at the same time. Instead, one should implement a
collection of rules belonging to one overall strategy. Although I can not tell, at present, how many such strat-
egies there could be, three candidates are binary articulation, articulation enhancing phrasing, and articulation
enhancing metrical structure. If the relative IOIs are computed from some other rules (as in, e.g. Director
Musices [21]), indications of a reasonable articulation could be given by using an inverse relationship between
relative IOI and articulation, even without applying any particular articulation strategy.
In conclusion, even though musical articulation might be hard to model, it is well worth trying. The present
study suggests that performers often articulate in comprehensible ways, but it is difficult to tell which strategy
is to be chosen in a particular piece. Some of the strategies are mutually exclusive. Still the mean intraper-
former correlation is almost as high for articulation as for relative IOI, the variations of which have been
studied and modelled by many researchers. Articulation seems to be no more random than, say, phrasing, but it
is certainly more complex.
6. REFERENCES
[1] Forsblom, E., Studier över stiltrohet och subjektivitet i interpretationen av J. S. Bachs orgelkomposi-
tioner, Acta Academiae Aboensis, Humaniora XXIII.1, Åbo, 1957, p. 54.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-23
[2] Ferguson, H., Keyboard Interpretation from the 14th to the 19th Century: An Introduction, Oxford Univer-
sity Press, New York & London, 1975, p. 53.
[3] Chew, G., “Articulation and phrasing,” in S. Sadie (ed.), The New Grove Dictionary of Music and Musi-
cians, vol. 2, 86–9, Macmillan, London, 2001.
[4] Juslin, P. N., “Communicating Emotion in Music Performance: A Review and Theoretical Framework,”
in P. N. Juslin & J. A. Sloboda (eds.), Music and Emotion: Theory and Research, 309–37, Oxford Uni-
versity Press, Oxford, 2001.
[5] Keller, H., Die musikalische Artikulation insbesondere bei Joh. Seb. Bach, Veröffentlichungen des
Musik-Instituts der Universität Tübingen, C. L. Schultheiß Musikverlag, Stuttgart, 1925.
[6] Keller, H., Phrasing and Articulation: A Contribution to a Rhetoric of Music. Translated by L. Gerdine.
Barrie and Rockliff, London, 1965, p. 4.
[7] Friberg, A., and Battel, G. U., “Structural Communication,” in R. Parncutt & G. E. McPherson (eds.), The
Science and Psychology of Music Performance: Creative Strategies for Teaching and Learning, 199–
218. Oxford University Press, New York, 2002.
[8] Bresin, R., and Battel, G. U., “Articulation Strategies in Expressive Piano Performance. Analysis of
Legato, Staccato, and Repeated Notes in Performance of the Andante Movement of Mozart’s Sonata in G
Major (K 545),” Journal of New Music Research 29(3), 211–24, 2000.
[9] Bolzinger, S., Contribution à l’étude de la rétroaction dans la pratique musicale par l’analyse de l’influ-
ence des variations d’acoustique de la salle sur le jeu du pianiste. Diss., Institut de Mécanique, Univer-
sité Aix-Marseille II, Marseille, 1995. (Exemplaire provisoire)
[10] Gabrielsson, A., “Once Again: The Theme from Mozart’s Piano Sonata in A Major (K. 331),” in A. Ga-
brielsson (ed.), Action and Perception in Rhythm and Music, Publications issued by the Royal Swedish
Academy of Music No 55, 81–103, 1987.
[11] Repp, B. H., “Diversity and commonality in music performance: An analysis of timing microstructure in
Schumann’s ‘Träumerei’,” Journal of the Acoustical Society of America 92(5), 2546–68, 1992.
[12] Repp, B. H., “Acoustics, Perception, and Production of Legato Articulation on a Digital Piano,” Journal
of the Acoustical Society of America 97(6), 3862–74, 1995.
[13] Repp, B. H., “Acoustics, Perception, and Production of Legato Articulation on a Computer-Controlled
Grand Piano,” Journal of the Acoustical Society of America 102(3), 1878–90, 1997.
[14] Repp, B. H., “Perception and Production of Staccato Articulation on the Piano,” unpublished manuscript,
completed October 1998. Available at http://www.haskins.yale.edu/haskins/MISC/REPP/Staccato.pdf.
[15] Battel, G. U., and Fimbianti, R., “How Communicate Expressive Intentions in Piano Performance,” Pro-
ceedings of the XII Colloquium on Musical Informatics, Gorizia, September 24-26, 1998, 67–70.
[16] Gabrielsson, A., “The Performance of Music,” in D. Deutsch, The Psychology of Music, Academic Press
San Diego/London, 2nd ed., 501–602, 1999.
[17] Gabrielsson, A., “Music Performance Research at the Millennium,” Psychology of Music 31(3), 221–72,
2003.
[18] Jerkert, J., “Measurements and models of musical articulation,” Proceedings of SMAC03 (Stockholm
Music Acoustics Conference 2003), 537–41, 2003.
[19] WaveSurfer 1.4.5, developed by K. Sjölander and J. Beskow, 2002. Available at http://www.speech.
kth.se/wavesurfer/.
[20] Soundswell 4.0, Hitech Development, 2002. Information available at http://www.hitech.se/development/.
[21] Director Musices 2.4, 2003, developed mainly by Anders Friberg, Lars Frydén, Vittorio Colombo, and
Johan Sundberg. Available free of charge at http://www.speech.kth.se/music/performance/.
Joint Baltic-Nordic Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland BNAM2004-24